---
title: "A stronger writing model, with automatic ..."
description: "Drafting now runs on a stronger primary model and automatically falls through two backup models, so a single AI provider outage no longer stops your campaign..."
canonical: "https://firstsales.io/changelog/stronger-writing-model-automatic-backups/"
---

1. [Home](/)
2. [Changelog](/changelog/)
3. A stronger writing model, with automatic backups

[All updates](/changelog/)

ImprovementAugust 6, 20265 min read

# A stronger writing model, with automatic backups

Drafting now runs on a stronger primary model and automatically falls through two backup models, so a single AI provider outage no longer stops your campaigns from drafting.

Drafting now runs on a stronger primary writing model, and if that model's provider has an outage, FirstSales automatically falls through two backup models so your campaigns keep drafting.

## What changed

FirstSales switched the model that writes your drafts to a stronger one as the default. On its own, that means better drafts: more natural phrasing, better handling of the instructions you set in Teach and the Learning tab, and fewer drafts that need manual correction before they're ready to send.

On top of the model change, drafting is no longer dependent on a single AI provider staying available. Previously, if the provider behind your drafting model had an outage or a slowdown, drafting for your campaigns could stall until that provider recovered, since there was nothing else for the system to fall back to. Now, if the primary model is unavailable, the system automatically tries a second model, and if that one is also unavailable, a third. Each fallback is ordered, so the system always tries the strongest available option first rather than picking at random.

This matters because your campaigns used to be exposed to whatever happened at whichever single provider was behind the drafting model at the time. An outage or a capacity issue at that provider meant your drafts stopped generating, your sequence fell behind schedule, and there was nothing to do but wait it out. With two fallback models in place, a single provider's problems no longer have to become your problem.

## How to use it

There's nothing to configure. Drafting uses the new primary model by default, and the fallback happens automatically in the background if that model becomes unavailable. You won't see which specific model produced a given draft, since the point of the fallback chain is that it shouldn't matter to you; you'll just see that drafts keep coming even during a provider disruption that would previously have stopped them.

If you notice a change in draft tone or quality on a particular day, it's worth checking your Learning tab and writing style settings first, since those still control the substance of what gets written regardless of which model in the fallback chain actually produced it. The fallback exists to keep drafting running, not to change what you've asked the AI to do.

## Why it matters

A campaign that stops drafting mid-run because of a provider outage costs you real time, and that cost compounds. A prospect who should have gotten a follow-up today doesn't get one until the outage clears, which pushes your whole sequence out of sync with the calendar you built it around, and can mean a slot-offer or a time-sensitive follow-up lands late enough to feel out of place when it finally arrives. None of that is under your control when it's tied to a single provider's uptime.

Moving to a stronger model as the default improves draft quality on its own terms, independent of any outage. The fallback chain solves a separate problem: it means an outage or capacity issue at one AI provider doesn't automatically become downtime for your outreach. Together, the two changes mean you get better drafts most of the time, and you keep getting drafts at all during the times you'd otherwise have been stuck waiting.

It's also worth being clear about what this change doesn't touch. Your instructions in Describe, Teach, and the Learning tab still fully control what your drafts say and how they're written. The model change and the fallback chain sit underneath that layer, handling which underlying system actually produces the words, not what those words are supposed to say. You should notice better default phrasing where your own instructions leave room for it, and you shouldn't notice anything at all when a fallback kicks in during an outage, since the goal of the fallback is precisely that it stays invisible to you.

## Troubleshooting a stalled campaign

There's no dashboard or setting that shows fallback activity, by design. If a campaign that was drafting normally suddenly stops, that's more likely a sign of a different issue worth checking directly, a paused campaign, an exhausted research or send budget, an instruction that's holding drafts for review, rather than something the fallback chain would have caught, since its entire purpose is to prevent exactly that kind of stall.

If you're comparing drafts from before and after this change and notice a real shift in quality, that's expected and is the primary model upgrade doing its job. Judge it the same way you'd judge any draft: against your own instructions and the Practice step, not against which model happened to write it.

## The cost of a missed cadence

For a business that depends on outreach going out on a predictable cadence, an unplanned pause is rarely just an inconvenience. Prospects who expected a follow-up and didn't get one can read the silence as disinterest, and a sequence that falls behind schedule during an outage doesn't automatically catch back up once the provider recovers, since some of those planned touches have simply expired. Removing single-provider dependency from the drafting pipeline takes one whole category of that risk off the table.

## A note on how the fallback is ordered

The fallback chain isn't a random pool of backup models the system reaches into whenever the primary is slow. It's an ordered sequence: try the strongest available model first, and only move to the next one if that attempt genuinely fails rather than just being slower than usual. That distinction matters for quality. A naive fallback that treated any delay as a reason to switch models could end up routing routine drafts through a weaker model unnecessarily; an ordered fallback that only activates on an actual failure keeps you on the best available option as often as possible, and only steps down when there's no other choice.

It's also worth stating what a fallback like this can and can't fix. It protects you from a specific kind of failure: the primary model becoming unavailable at the provider level. It doesn't change anything about your own campaign settings, your instructions, or how much research is available for a given contact, since those are independent of which model actually writes the words. If drafting stops for a reason unrelated to model availability, the fallback chain won't be what resolves it, and the underlying campaign status is still the right place to look first.

## Availability

This applies to all campaigns using AI-generated drafts, across first-touch, follow-up, and reply emails.

[ PreviousDrafts read like a person wrote them](/changelog/drafts-read-like-a-person-wrote-them/)[Next No outreach goes out on guesswork](/changelog/no-outreach-goes-out-on-guesswork/)